Added text summary files and .gitignore

.gitignore

@@ -0,0 +1,38 @@
+# Compiled Object files
+*.slo
+*.lo
+*.o
+*.obj
+
+# Precompiled Headers
+*.gch
+*.pch
+
+# Compiled Dynamic libraries
+*.so
+*.dylib
+*.dll
+
+# Fortran module files
+*.mod
+
+# Compiled Static libraries
+*.lai
+*.la
+*.a
+*.lib
+
+# Executables
+*.exe
+*.out
+*.app
+
+#
+*.dat
+*~
+gp
+gplvm
+ivm
+*.gp
+*.model
+octave-workspace

CClctrl.txt

@@ -0,0 +1,8 @@
+This file provides a base class for command line control. Various functionality such as parsing arguments and reading input files is provided.
+
+File Notes
+
+21/10/2005 Incorporated William Baxter's isCurrentArg modification.
+
+20/10/2005 Minor modifications by William V. Baxter as part of the porting to Microsoft Visual C++.
+

CIvm.txt

@@ -0,0 +1,5 @@
+This file contains classes for representing Gaussian processes using the IVM. The model relies on the CKern class and the CNoise class to define the kernel and the noise model.
+
+01/03/2005 Gregg Keller had problems loading in a data set. It turned out it was when the last feature in an active point was 0 no end of line character was placed on the file. This is fixed as part of release 0.12.
+
+23/10/2005 Modifications for compiling under Visual C++, by Neil.

CKern.txt

@@ -0,0 +1,9 @@
+This file contains classes for representing kernels. These classes do not store the kernel matrix, they merely store the kernels characteristics. There are several types of kernel provided, and it should be easy to add others. CCmpndKern provides a class which forms an additive kernel from individual kernels (i.e. a linear plus an RBF kernel).
+
+16/03/2007 Added code to compute parameter gradients when two X matrices are provided. This enables use of the toolbox with DTC, FITC, PITC approximations. Also changed the transformation of kernel parameters to be by exponential rather than negLogLogit, in line with the default in the MATLAB kern toolbox.
+
+22/10/2005 Added updateX method to class, from William V. Baxter's changes to make computation more efficient. The changes involve precomputing the distance matrix and a portion of the kernel matrix and storing it for later use. UpdateX is called when X has changed so that these stored values are recomputed.
+
+21/10/2005 More changes from William V. Baxter. Most are removal of redundant const modifiers, but also vector<CMatrix*> is now being passed as a reference, rather than by value.
+
+20/10/2005 Incorporated changes from William V. Baxter to allow compilation under MSVC. 

CMatrix.txt

@@ -0,0 +1,5 @@
+This file contains the basic matrix class. The matrix hierarchy is currently very flat and is designed to be a simple (and hopefully efficient) interface to the LAPACK and BLAS libraries. 
+
+21/10/2005 More changes from William Baxter. In particular more redundant consts are removed and a symvColColOff is added for offset multiplication. Some bugs are removed (including in sumCol) and casting is made clearer in many places. Also in several places vector<> is now used instead of declaring an array.
+
+20/10/2005 Incorporated William Baxter's changes which allow compilation of files under MSVC. These included a number of bug fixes (such as my embarrassing use of otherwise: rather than default:!). Also included is a new Matrix method addVal, which is a sensible way of adding to an element.

CMltools.txt

@@ -0,0 +1 @@
+This file contains classes for general machine learning models such as multi-layer perceptrons.

CNoise.txt

@@ -0,0 +1,11 @@
+This file contains classes for different noise models including classification, regression and semi-supervised learning. 
+
+10/10/2006 Fixed sign error in likelihood for ordered model. This error was propagated because of a sign error in ndlutil.cpp. In this case two wrongs made a right.
+
+1/3/2005 While looking for the bug that caused Gregg Keller problems, I found that when noise models where read from disk they were allocating storeage at the rong point. Moved initStoreage() out of setNumProcesses and set the numData to a dummy value in readParamsFromStream() to fix it.
+
+This file now also contains the code for ordered categorical noise models.
+
+21/10/2005 More changes from Bill, mostly removing redundant const modifiers.
+
+20/10/2005 Incorporated William V. Baxter's code modifications for running the code with MSVC. Mostly this involved removing redundant const statements.

CTransform.txt

@@ -0,0 +1,7 @@
+This file contains classes that enable non-linear transformation of parameters. This is so that in the space where these parameters are optimised they can be optimised in a unconstrained way, but in the real parameter space they are effectively constrained. The `right' way to do this is through constrained optimisation, but this approach allows for the use of simple, effective algorithms such as scaled conjugate gradient. One commonly used transformation is a log transformation of a positive parameter. 
+
+16/3/2007 Added CExpTransform for performing logarithmic transformations of kernel parameters in line with the new default in the MATLAB optimi toolbox.
+
+21/10/2005 More minor changes to do with removing redundant const modifiers.
+
+20/10/2005 Changes from William V. Baxter for compilation under MSVC. The psoition of a couple of declarations has been changed and redundant consts have been removed.